Contents of Volume 22 (2012)

1/2012 2/2012 3/2012 4/2012 5/2012

6/2012

  • [1] Yalkin C., Korkmaz E. E. (Turkey):
    Neural network world: A neural network based selection method for genetic algorithms, 495-510.

    Full text     DOI: 10.14311/NNW.2012.22.030

    Genetic algorithms (GAs) are stochastic methods that are widely used in search and optimization. The breeding process is the main driving mechanism for GAs that leads the way to find the global optimum. And the initial phase of the breeding process starts with parent selection. The selection utilized in a GA is effective on the convergence speed of the algorithm. A GA can use different selection mechanisms for choosing parents from the population and in many applications the process generally depends on the fitness values of the individuals. Artificial neural networks (ANNs) are used to decide the appropriate parents by the new hybrid algorithm proposed in this study. And the use of neural networks aims to produce better offspring during the GA search. The neural network utilized in this algorithm tries to learn the structural patterns and correlations that enable two parents to produce high-fit offspring. In the breeding process, the first parent is selected based on the fitness value as usual. Then it is the neural network that decides the appropriate mate for the first parent chosen. Hence, the selection mechanism is not solely dependent on the fitness values in this study. The algorithm is tested with seven benchmark functions. It is observed from results of these tests that the new selection method leads genetic algorithm to converge faster.


  • [2] Bostik J., Kukal J., Virius M. (Czech Republic):
    Perceptron network for RBF lovers, 511-518.

    Full text     DOI: 10.14311/NNW.2012.22.031

    There are two basic types of artificial neural networks: Multi-Layer Perceptron (MLP) and Radial Basis Function network (RBF). The first type (MLP) consists of one type of neuron, which can be decomposed into a linear and sigmoid part. The second type (RBF) consists of two types of neurons: radial and linear ones. The radial basis function is analyzed and then used for decomposition of RBF network. The resulting Perceptron Radial Basis Function Network (PRBF) consists of two types of neurons: linear and extended sigmoid ones. Any RBF network can be directly converted to a four-layer PRBF network while any MLP network with three layers can be approximated by a five-layer PRBF network. The new PRBF network is then a generalization of MLP and RBF network abilities. Learning strategies are also discussed. The new type of PRBF network and its learning via repeated local optimization is demonstrated on a numerical example together with RBF and MLP for comparison. This paper is organized as follows: Basic properties of MLP and RBF neurons are summarized in the first two chapters. The third chapter includes novel relationship between sigmoidal and radial functions, which is useful for RBF decomposition and generalization. Description of new PRBF network, together with its properties, is subject of the fourth chapter. Numerical experiments with a PRBF and their requests are given in the last chapters.


  • [3] Bártů M. (Czech Republic): Speech disorder analysis using Matching Pursuit and Kohonen Self-Organizing Maps, 519-533.

    Full text     DOI: 10.14311/NNW.2012.22.032

    The method described in the following text was developed to analyze disordered children speech. The diagnosis of the children is developmental dysphasia. Since developmental dysphasia has impact on children's speech ability, the classification of utterances helps to determine whether treatment and medication are appropriate. The paper describes the method developed to provide classification based on utterances but without any additional demands on speech preprocessing (e.g. labeling). The method uses Matching Pursuit algorithm for speech parameterization and Kohonen Self-Organizing Maps for extraction of features from utterances. Features extracted from the utterances of healthy children are then compared to features obtained from the speech of children suffering from the illness.


  • [4] Babak M. A., Sharafat A. R., Setarehdan S. K. (Iran): An adaptive backpropagation neural network for arrhythmia classification using R-R interval signal, 535-548.

    Full text     DOI: 10.14311/NNW.2012.22.033

    Automatic detection and classification of cardiac arrhythmias with high accuracy and by using as little information as possible is highly useful in Holter monitoring of the high risk patients and in telemedicine applications where the amount of information which must be transmitted is an important issue. To this end, we have used an adaptive-learning-rate neural network for automatic classification of four types of cardiac arrhythmia. In doing so, we have employed a mix of linear, nonlinear, and chaotic features of the R-R interval signal to significantly reduce the required information needed for analysis, and substantially improve the accuracy, as compared to existing systems (both ECG-based and R-R interval-based). For normal sinus rhythm (NSR), premature ventricular contraction (PVC), ventricular fibrillation (VF), and atrial fibrillation (AF), the discrimination accuracies of 99.59%, 99.32%, 99.73%, and 98.69% were obtained, respectively on the MIT-BIH database, which are superior to all existing classifiers.


  • [5] Hongjiu L., Rieg R., Yanrong H. (P. R. China): Performance comparison of artificial intelligence methods for predicting cash flow, 549-564.

    Full text     DOI: 10.14311/NNW.2012.22.034

    Cash flow forecasting is indispensable for managers, investors and banks. However, which method is more robust has been argued under the condition of small size samples. With sliding window technique we create the Response Surface, Back Propagation Neural Network, Radial Basis Functions Neural Network and Support Vector Machine models respectively, which are examined by comparing performances of training and simulation. Performances of training models are measured by mean of squared errors while that of simulation is done by average relative errors of the results. By comparison, Support Vector Machine is most robust to forecast cash flow, followed by Radial Basis Function Neural Network, the third Back Propagation Neural Network and the last Response Surface Model. The optimal result of each model depends on the window size of the transmitter.


  • [6] Frolov A., Húsek D., Polyakov P. Y., Řezanková H. (Czech Republic, Russia): A comparative study of two methodologies for binary datasets analysis, 565-582.

    Full text     DOI: 10.14311/NNW.2012.22.035

    Studied are differences of two approaches targeted to reveal latent variables in binary data. These approaches assume that the observed high dimensional data are driven by a small number of hidden binary sources combined due to Boolean superposition. The first approach is the Boolean matrix factorization (BMF) and the second one is the Boolean factor analysis (BFA). The two BMF methods are used for comparison. First is the M8 method from the BMDP statistical software package and the second one is the method suggested by Belohlavek \& Vychodil. These two are compared to BFA, especially with the Expectation-maximization Boolean Factor Analysis we had developed earlier has, however, been extended with a binarization step developed here. The well-known bars problem and the mushroom dataset are used for revealing the methods' peculiarities. In particular, the reconstruction ability of the computed factors and the information gain as the measure of dimension reduction was under scrutiny. It was shown that BFA slightly loses to BMF in performance when noise-free signals are analyzed. Conversely, BMF loses considerably to BFA when input signals are noisy.


  • [7] Contents volume 22 (2012), 583-586.

  • [8] Author's index volume 22 (2012), 587-589.

5/2012

  • [1] Zainuddin Z., Ong P. (Malaysia): An effective and novel wavelet neural network approach in classifying type 2 diabetics,407-428.

    Full text     DOI: 10.14311/NNW.2012.22.025

    Designing a wavelet neural network (WNN) needs to be done judiciously in attaining the optimal generalization performance. Its prediction competence relies highly on the initial value of translation vectors. However, there is no established solution in determining the appropriate initial value for the translation vectors at this moment. In this paper, we propose a novel enhanced fuzzy c-means clustering algorithm - specifically, the modified point symmetry-based fuzzy c-means (MPSDFCM) algorithm - in initializing the translation vectors of the WNNs. The effectiveness of embedding different activation functions in WNNs will be investigated as well. The categorization effectiveness of the proposed WNNs model was then evaluated in classifying the type 2 diabetics, and was compared with the multilayer perceptrons (MLPs) and radial basis function neural networks (RBFNNs) models. Performance assessment shows that our proposed model outperforms the rest, since a 100% superior classification rate was achieved.


  • [2] Pobar M., Martinčić-Ipšić S., Ipšić I. (Croatia): Optimization of cost function weights for unit selection speech synthesis using speech recognition,429-441.

    Full text     DOI: 10.14311/NNW.2012.22.026

    A well known problem in unit selection speech synthesis is designing the join and target function sub-costs and optimizing their corresponding weights so that they reflect the human listeners' preferences. To achieve this we propose a procedure where an objective criterion for optimal speech unit selection is used. The objective criterion for tuning the cost function weights is based on automatic speech recognition results. In order to demonstrate the effectiveness of the proposed method listening tests with 31 naive listeners were performed. The experimental results have shown that the proposed method improves speech quality and intelligibility. In order to evaluate the quality of synthesized speech the unit selection speech synthesis system is compared with two other Croatian speech synthesis systems with voices built using the same recorded speech corpus. One of these voices was built with the Festival speech synthesis system using the statistical parametric method and the other is a diphone concatenation based text-to-speech system. The comparison is based on subjective tests using MOS (mean opinion score) evaluation. The system using the proposed method used for cost function weights optimization performs better than other compared systems according to the subjective tests.


  • [3] Arslan M. H., Ceylan M., Koyuncu T. (Turkey): An ANN approaches on estimating earthquake performances of existing RC buildings, 443-458.

    Full text     DOI: 10.14311/NNW.2012.22.027

    This study aims at developing an artificial intelligence-based (ANN based) analytical method to analyze earthquake performances of the reinforced concrete (RC) buildings. In the scope of the present study, 66 real RC buildings with four to ten storeys were subject to performance analysis according to 19 parameters considered effective on the performance of RC buildings. In addition, the level of performance of these buildings in case of an earthquake was determined on the basis of the 4-grade performance levels specified in Turkish Earthquake Code-2007 (TEC-2007). Thus, an output performance data group was created for the analyzed buildings, in accordance with the input data. Thanks to the ANN-based fast evaluation algorithm mentioned above and developed within the scope of the proposed project study, it will be possible to make an economic and rapid evaluation of four to ten-storey RC buildings in Turkey with great accuracy (about 80%). Detection of post-earthquake performances of RC buildings in the scope of the present study will facilitate reaching important results in terms of buildings, which will be beneficial for Civil Engineers of Turkey and similar countries.


  • [4] Chien-Hsing Chen, Chung-Chian Hsu (Taiwan): Using text and visual mining to analyze clinical diagnosis records 459-478.

    Full text     DOI: 10.14311/NNW.2012.22.028

    Hospitals must index each case of inpatient medical care with codes from the International Classification of Diseases, 9th Revision (ICD-9), under regulations from the Bureau of National Health Insurance. This paper aims to investigate the analysis of free-textual clinical medical diagnosis documents with ICD-9 codes using state-of-the-art techniques from text and visual mining fields. In this paper, ViSOM and SOM approaches inspire several analyses of clinical diagnosis records with ICD-9 codes. ViSOM and SOM are also used to obtain interesting patterns that have not been discovered with traditional, nonvisual approaches. Furthermore, we addressed three principles that can be used to help clinical doctors analyze diagnosis records effectively using the ViSOM and SOM approaches. The experiments were conducted using real diagnosis records and show that ViSOM and SOM are helpful for organizational decision-making activities.


  • [5] Demir G. K., Dural M. U., Alyuruk H., Cavas L. (Turkey): Artificial neural network model for biosorption of methylene blue by dead leaves of Posidonia oceanica (L.) Delile 479-494.

    Full text     DOI: 10.14311/NNW.2012.22.029

    In the present study, an alternative promising evaluation method was recommended for dead leaves of Posidonia oceanica (L.) Delile as an adsorbent for biosorption of Methylene Blue (MB). The data from batch experiments were modeled by using Artificial Neural Network (ANN). The optimal operation conditions for biosorption of MB by P. oceanica dead leaves were found for pH, adsorbent dosage, temperature and initial dye concentration as 6, 0.3 g, 303 K and 50 mg/L, respectively. The adsorption reached equilibrium after 30 minutes. According to the results of sensitivity analysis, relative importance of temperature, dye concentration, pH, adsorbent dosage and process time on the biosorption of MB were 33%, 27%, 21%, 10% and 8%, respectively. Minimum mean square error (MSE) was found as 0.0169 by ANN modeling. The present study reveals a novel strategy for adsorption studies to utilize the highly accumulated biomass of dead leaves of P. oceanica in Turkish coastlines instead of burning these dead leaves.


4/2012

  • [1] Yajun Du, Yingyu Wang, Shaoming Chen (China): The understanding between agent crawlers based on domain ontology, 311-324.

    Full text     DOI: 10.14311/NNW.2012.22.018

    Ontology is widely used in the computer domain to structure concepts that represent a view of world nowadays, which could formally specify semantic relationship among the terms. In this paper, we present coordination between agent crawlers based on ontology in Topic Specific Search Engines, and we try to measure understanding among them, relying on Formal Concept Analysis (FCA) instead of comparing the terms only. In literature, most papers on concept similarity in FCA are based on two different concepts in the same concept lattice, and whereas there is very little research related to different concept lattices or even different agents. We propose a novel method on concept similarity for computing the Concept-Concept similarity, the Concept-Ontology similarity and the Ontology-Ontology similarity, and at last we can deduce understanding among agent crawlers. Finally, we can guide the crawlers effectively in our Search Engine.


  • [2] Panchi Li, Haiying Wang (China): Quantum ant colony optimization algorithm based on Bloch spherical search, 325-341.

    Full text     DOI: 10.14311/NNW.2012.22.019

    In the existing quantum-behaved optimization algorithms, almost all of the individuals are encoded by qubits described on plane unit circle. As qubits contain only a variable parameter, quantum properties have not been fully embodied, which limits the optimization ability rise further. In order to solve this problem, this paper proposes a quantum ant colony optimization algorithm based on Bloch sphere search. In the proposed algorithm, the positions of ants are encoded by qubits described on Bloch sphere. First, the destination to move is determined according to the select probability constructed by the pheromone and heuristic information, then, the rotation axis is established with Pauli matrixes, and the evolution search is realized with the rotation of qubits on Bloch sphere. In order to avoid premature convergence, the mutation is performed with Hadamard gates. Finally, the pheromone and the heuristic information are updated in the new positions of ants. As the optimization process is performed in $n$-dimensional hypercube space $[-1,1]^{n}$, which has nothing to do with the specific issues, hence, the proposed method has good adaptability for a variety of optimization problems. The simulation results show that the proposed algorithm is superior to other quantum-behaved optimization algorithms in both search ability and optimization efficiency.


  • [3] Chunguo Fei, Baili Su (China): Adding Decaying Self-feedback Continuous Hopfield Neural Network Convergence Analysis in the Hyper-cube Space, 343-355.

    Full text     DOI: 10.14311/NNW.2012.22.020

    After sigmoid activation function is replaced with piecewise linear activation function, the adding decaying self-feedback continuous Hopfield neural network (ADSCHNN) searching space changes to hyper-cube space, i.e. the simplified ADSCHNN is obtained. Then, convergence analysis is given for the simplified ADSCHNN in hyper-cube space. It is proved through convergence analysis that the ADSCHNN outperforms the continuous Hopfield neural network (CHNN), when they are applied to solve optimization problem. It is also proved that when extra self-feedback is negative, the ADSCHNN is more effective than the extra self-feedback is positive, when the ADSCHNN is applied to solve TSP.


  • [4] Lopes N., Ribeiro B. (Portugal): Handling Missing Values via a Neural Selective Input Model, 357-370.

    Full text     DOI: 10.14311/NNW.2012.22.021

    Missing data represent an ubiquitous problem with numerous and diverse causes. Handling Missing Value properly is a crucial issue, in particular in Machine Learning and pattern recognition. To date, the only option available for standard Neural Network to handle this problem has been to rely on pre-processing techniques such as imputation for estimating the missing data values, which limited considerably the scope of their application. To circumvent this limitation we propose a Neural Selective Input Model that accommodates different transparent and bound models, while providing support for Neural Network to handle Missing Value directly. By embedding the mechanisms to support Missing Value we can obtain better models that reflect the uncertainty caused by unknown values. Experiments on several UCI datasets with both different distributions and proportion of Missing Value show that the Neural Selective Input Model approach is very robust and yields good to excellent results. Furthermore, the Neural Selective Input Model performs better than the state-of-the-art imputation techniques either with higher prevalence of Missing Value in a large number of features or with a significant proportion of Missing Value, while delivering competitive performance in the remaining cases. We demonstrate the usefulness and validity of the Neural Selective Input Model, making this a first-class method for dealing with this problem.


  • [5] Martinovič J., Novosád T., Snášel V., Scherer P., Klement P., Šebesta R. (Czech Republic): Clustering the mobile phone positions based on suffix tree and Self-Organizing Maps, 371-386.

    Full text     DOI: 10.14311/NNW.2012.22.022

    In this article we present a novel method for mobile phone positioning using a vector space model, suffix trees and an information retrieval approach. The algorithm is based on a database of previous measurements which are used as an index which looks for the nearest neighbor toward the query measurement. The accuracy of the algorithm is, in most cases, good enough to accomplish the E9-1-1 standards requirements on tested data. In addition, we are trying to look at the clusters of patterns that we have created from measured data and we have reflected them to the map. We use Self-Organizing Maps for these purposes.


  • [6] Vlček M. (Czech Republic): Chebyshev polynomial approximation for activation sigmoid function, 387-393.

    Full text     DOI: 10.14311/NNW.2012.22.023

    An alternative polynomial approximation for the activation sigmoid function is developed here. It can considerably simplify the input/output operations of a neural network. The recursive algorithm is found for Chebyshev expansion of all constituting polynomials.


  • [7] Holota R. (Czech Republic): Colour image recognition based on single-layer neural networks of Min/Max nodes, 395-405.

    Full text     DOI: 10.14311/NNW.2012.22.024

    An image recognition system can be based on a single-layer neural network composed of Min/Max nodes. This principle is easy to use for greyscale images. However, this article deals with the possibilities of utilising neural nets for colour image recognition. Several principles are demonstrated and tested by recently developed software. A new modified Min/Max node Single Layer Net, suitable for recognition in HSV (Hue Saturation Value) colour space, is presented in this paper.


3/2012

  • [1] Tibebe Beshah, Dejene Ejigu, Ajith Abraham, Václav Snášel, Pavel Krömer (Ethiopia, Czech Republic): Knowledge discovery from road traffic accident data in Ethiopia: Data quality, ensembling and trend analysis for improving road safety, 215-244.

    Full text     DOI: 10.14311/NNW.2012.22.013

    Descriptive analysis of the magnitude and situation of road safety in general and road accidents in particular is important, but understanding of data quality, factors related with dangerous situations and various interesting patterns in data is of even greater importance. Under the umbrella of information architecture research for road safety in developing countries, the objective of this machine learning experimental research is to explore data quality issues, analyze trends and predict the role of road users on possible injury risks. The research employed TreeNet, Classification and Adaptive Regression Trees (CART), Random Forest (RF) and hybrid ensemble approach. To identify relevant patterns and illustrate the performance of the techniques for the road safety domain, road accident data collected from Addis Ababa Traffic Office is subject to several analyses. Empirical results illustrate that data quality is a major problem that needs architectural guideline and the prototype models could classify accidents with promising accuracy. In addition, an ensemble technique proves to be better in terms of predictive accuracy in the domain under study.


  • [2] Shichang Sun, Hongbo Liu, Pixi Zhao, Hongfei Lin (China): Two-stage model selection with parameters weighted hidden Markov models and likelihood ratio for part-of-speech tagging, 245-262.

    Full text     DOI: 10.14311/NNW.2012.22.014

    In many natural language processing applications two or more models usually have to be involved for accuracy. But it is difficult for minor models, such as “backoff” taggers in part-of-speech tagging, to cooperate smoothly with the major probabilistic model. We introduce a two-stage approach for model selection between hidden Markov models and other minor models. In the first stage, the major model is extended to give a set of candidates for model selection. Parameters weighted hidden Markov model is presented using weighted ratio to create the candidate set. In the second stage, heuristic rules and features are used as evaluation functions to give extra scores to candidates in the set. Such scores are calculated using a diagnostic likelihood ratio test based on sensitivity and specificity criteria. The selection procedure can be fulfilled using swarm optimization technique. Experiment results on public tagging data sets show the applicability of the proposed approach.


  • [3] Yao J. B., Yao B. Z., Li L., Jiang Y. L. (P.R.China): Hybrid model for displacement prediction of tunnel surrounding rock, 263-275.

    Full text     DOI: 10.14311/NNW.2012.22.015

    This paper presents a hybrid method to predict tunnel surrounding rock displacement, which is one of the most important factors for quality control and safety during tunnel construction. The hybrid method comprises two phases, one is support vector machine (SVM)-based model for predicting the tunnel surrounding rock displacement, and the other is GA-based model for optimizing the parameters in the SVM. The proposed model is evaluated with the data of tunnel surrounding rock displacement on the tunnel of Wuhan-Guangzhou railway in China. The results show that genetic algorithm (GA) has a good convergence and relative stable performance. The comparison results also show that the hybrid method can generally provide a better performance than artificial neural network (ANN) and finite element method (FEM) for tunnel surrounding rock displacement prediction.


  • [4] Petr Ježek, Roman Mouček (Czech Republic): System for EEG/ERP data and metadata storage and management, 277-290.

    Full text     DOI: 10.14311/NNW.2012.22.016

    The purpose of this paper is to introduce a system for EEG/ERP (electroencephalography, event-related potentials) data and metadata storage and processing and to summarize the authors' research in this field. Since researchers have difficulties with a~suitable long-term storage and management of electrophysiology data the presented system helps them to increase both efficiency and effectiveness of their work by providing the means for the storage, management, search and sharing of EEG/ERP data. The requirements specification including the system context, system requirements, project scope, basic features, system users, and data formats and metadata structures are presented. The database structure is proposed; upload, download and interchange of EEG/ERP data and metadata using the web interface are described. The system architecture, used technologies and the final realization are described. Data and metadata search over the system and user accounts including system security management are also presented. Additional tools and structures as converters of data formats and semantic web ontology are mentioned.


  • [5] Reza Ebrahimpour, Kioumars Babakhani, Seyed Ali Asghar Abbaszadeh Arani, Saeed Masoudnia (Iran): Epileptic seizure detection using a neural network ensemble method and wavelet transform, 291-310.

    Full text     DOI: 10.14311/NNW.2012.22.017

    This paper presents a new method to automate the process of epileptic seizure detection in electroencephalogram (EEG) signals using wavelet transform and an improved version of negative correlation learning (NCL) algorithm. An improved version of NCL is proposed by incorporating the capability of gating network, as a dynamic combining part of the mixture of experts (ME), into the combining outputs of base experts which are trained using negative correlation learning algorithm. The NCL training algorithm encourages the base experts to learn different parts or aspects of data set and the gating network provides the local competence of these base experts. Three types of normal (recorded from five healthy persons with eyes open), seizure-free (recorded from epileptogenic zoon of five patients) and epileptic EEG signals were decomposed into wavelet coefficients using discrete wavelet transform. Then the statistical features of the wavelet coefficients were computed representing them into the classifiers. Experimental results show that our proposed method classifies normal, seizure-free and epileptic EEG signals with the accuracy of 96.92% which is significantly better than previous combining methods.


2/2012

  • [1] Švantner J., Farkaš I., Crocker M. (Slovakia, Germany): Modeling utterance-driven visual attention during situated comprehension, 85-101.

    Full text     DOI: 10.14311/NNW.2012.22.006

    Evidence from behavioral studies demonstrates that spoken language guides attention in a related visual scene and that attended scene information can influence the comprehension process. Here we model sentence comprehension within visual contexts. A recurrent neural network is trained to associate the linguistic input with the visual scene and to produce the interpretation of the described event which is part of the visual scene. A feedback mechanism is investigated, which enables explicit utterance-mediated attention shifts to the relevant part of the scene. We compare four models - a simple recurrent network (SRN) and three models with specific types of additional feedback - in order to explore the role of the attention mechanism in the comprehension process. The results show that all networks learn not only successfully to produce the interpretation at the sentence end, but also demonstrate predictive behavior reflected by the ability to anticipate upcoming constituents. The SRN performs expectedly very well, but demonstrates that adding an explicit attentional mechanism does not lead to loss of performance, and even results in a slight improvement in one of the models.


  • [2] Kudelka M., Horak Z., Vozenilek V., Snasel V. (Czech Republic): Orthophoto feature extraction and clustering, 103-121.

    Full text     DOI: 10.14311/NNW.2012.22.007

    In this article we use a combination of neural networks with other techniques for the analysis of orthophotos. Our goal is to obtain results that can serve as a useful groundwork for interactive exploration of the terrain in detail. In our approach we split an aerial photo into a regular grid of segments and for each segment we detect a set of features. These features depict the segment from the viewpoint of a general image analysis (color, tint, etc.) as well as from the viewpoint of the shapes in the segment. We perform clustering based on the Formal Concept Analysis (FCA) and Non-negative Matrix Factorization (NMF) methods and project the results using effective visualization techniques back to the aerial photo. The FCA as a tool allows users to be involved in the exploration of particular clusters by navigation in the space of clusters. In this article we also present two of our own computer systems that support the process of the validation of extracted features using a neural network and also the process of navigation in clusters. Despite the fact that in our approach we use only general properties of images, the results of our experiments demonstrate the usefulness of our approach and the potential for further development.


  • [3] Veselý A. (Czech Republic): Modeling deduction with recurrent neural networks, 123-137.

    Full text     DOI: 10.14311/NNW.2012.22.008

    In the paper, we focus on reasoning with IF-THEN rules in propositional fragment of predicate calculus and on its modeling with neural networks. At first, IF-THEN deduction from facts is defined. Then it is proved that for any non-contradictory set of IF-THEN rules and literals (representing facts) there exists a layered recurrent network with 2 hidden layers that can specify all IF-THEN deducible literals. If we denote the set of all literal IF-THEN consequences as D_0 and the set of all literal logical consequences as D, then obviously D_0 \subset D. Thus, D_0 can be considered to be an approximation of D. Using the designed network for simulation of contradiction proof, the approximation D_0 may be easily refined. Furthermore, the network may also be used for determination of D. However, the algorithm that realizes necessary network computations has exponential complexity.


  • [4] Vintr T., Vintrova V., Rezankova H. (Czech Republic): Poisson distribution based initialization for fuzzy clustering, 139-159.

    Full text     DOI: 10.14311/NNW.2012.22.009

    A quality of centroid-based clustering is highly dependent on initialization. In the article we propose initialization based on the probability of finding objects, which could represent individual clusters. We present results of experiments which compare the quality of clustering obtained by k-means algorithm and by selected methods for fuzzy clustering: FCM (fuzzy c-means), PCA (possibilistic clustering algorithm) and UPFC (unsupervised possibilistic fuzzy clustering) with different initializations. These experiments demonstrate an improvement in the quality of clustering when initialized by the proposed method. The concept how to estimate a ratio of added noise is also presented.


  • [5] Beňušková L., Jedlička P. (New Zealand, Germany): Computational modeling of long-term depression of synaptic weights: insights from STDP, metaplasticity and spontaneous activity, 161-180.

    Full text     DOI: 10.14311/NNW.2012.22.010

    Using the STDP rule with metaplasticity, we show that to evoke long-term depression (LTD) or depotentiation of synaptic weights in the spiking model of granule cell is not easy. This is in accordance with a number of experimental studies. On the other hand, heterosynaptic LTD which accompanies homosynaptic long-term potentiation (LTP) is induced readily both in the model as well as in experiments. We offer possible explanation of these phenomena from STDP, metaplasticity and spontaneous activity. We suggest conditions under which it would be possible to induce homosynaptic LTD and depotentiation.


  • [6] Brandejsky T. (Czech Republic): Evolutionary system to model structure and parameters regression, 181-194.

    Full text     DOI: 10.14311/NNW.2012.22.011

    This paper discusses features of multilayered evolutionary system suitable to identify various systems including their model symbolic regression. Improved sensitivity allows modeling of difficult systems as deterministic chaos ones. The presented paper starts with a brief introduction to previous works and ideas which allowed to build the presented two abstraction levels system. Then the structure of Genetic Programming Algorithm - Evolutionary Strategy hybrid system is described and analyzed, including such problems as suitability to parallel implementation, optimal set of building blocks, or initial population generating rules. GPA-ES system combines GPA to model development with ES used for model parameter estimation and optimization. Such a hybrid system eliminates many weaknesses of standard GPA. The paper concludes with examples of GPA-ES application to Lorenz and Rősler systems regression and suggests application to Neural Network Model design.


  • [7] Anilkumar G. K. (Thailand): The subjective job scheduler with a satisfying criterion based on backpropagation neural network, 195-213.

    Full text     DOI: 10.14311/NNW.2012.22.012

    This paper aims to present and discuss the concept of a subjective job scheduler with a satisfying criterion based on a Backpropagation Neural Network (BPNN) and a greedy task alignment procedure. The BPNN is to assign priorities to the tasks of each job based on the given subjective criteria. The subjective criteria and the task alignment procedure depend on the solution plan towards a given job scheduling problem depending on the user's need. When the scheduler is provided with a desired job selection criteria and task alignment procedure for the problem, it generates user satisfying solutions for a set of jobs. The satisfying criterion of the scheduler determines the user satisfaction based on three measures: convergence test of the BPNN, validity of the input job set and cost evaluation of the solutions. The simulations and comparisons presented in this paper indicate that the proposed approach is one of the most effective strategies of structuring a subjective functional job scheduler.


1/2012

  • [1] Editorial, 1-2.
  • [2] Sabeti M., Boostani R., Zoughi T. (Iran): Using genetic programming to select the informative EEG-based features to distinguish schizophrenic patients, 3-20.

    Full text     DOI: 10.14311/NNW.2012.22.001

    There is growing interest to analyze electroencephalogram (EEG) signals with the objective of classifying schizophrenic patients from the control subjects. In this study, EEG signals of 15 schizophrenic patients and 19 age-matched control subjects are recorded using twenty surface electrodes. After the preprocessing phase, several features including autoregressive (AR) model coefficients, band power and fractal dimension were extracted from their recorded signals. Three classifiers including Linear Discriminant Analysis (LDA), Multi-LDA (MLDA) and Adaptive Boosting (Adaboost) were implemented to classify the EEG features of schizophrenic and normal subjects. Leave-one (participant)-out cross validation is performed in the training phase and finally in the test phase; the results of applying the LDA, MLDA and Adaboost respectively provided 78%, 81% and 82% classification accuracies between the two groups. For further improvement, Genetic Programming (GP) is employed to select more informative features and remove the redundant ones. After applying GP on the feature vectors, the results are remarkably improved so that the classification rate of the two groups with LDA, MLDA and Adaboost classifiers yielded 82%, 84% and 93% accuracies, respectively.


  • [3] Frolov A., Húsek D., Bobrov P., Korshakov A., Chernikova L., Konovalov R., Mokienko O. (Czech Republic, Russia): Sources of EEG activity most relevant to performance of brain-computer interface based on motor imagery, 21-37.

    Full text     DOI: 10.14311/NNW.2012.22.002

    The paper examines sources of brain activity, contributing to EEG patterns which correspond to motor imagery during training to control brain-computer interface. To identify individual source contribution into electroencephalogram recorded during the training Independent Component Analysis was used. Then those independent components for which the BCI system classification accuracy was at maximum were treated as relevant to performing the motor imagery tasks, since they demonstrated well exposed event related de-synchronization and event related synchronization of the sensorimotor μ-rhythm during imagining of contra- and ipsilateral hand movements. To reveal neurophysiological nature of these components we have solved the inverse EEG problem to locate the sources of brain activity causing these components to appear in EEG. The sources were located in hand representation areas of the primary sensorimotor cortex. Their positions practically coincide with the regions of brain activity during the motor imagination obtained in fMRI study. Individual geometry of brain and its covers provided by anatomical MR images was taken into account when localizing the sources.


  • [4] Kırlangıc A., Bacak-Turan G. (Turkey): On the rupture degree of a graph, 39-51.

    Full text     DOI: 10.14311/NNW.2012.22.003

    In a communication network, vulnerability measures the resistance of the network to disruption of operation after the failure of certain stations or communication links. If we think of a connected graph as model-ing a network, the rupture degree of a graph is one measure of graph vulnerability and it is defined by


    r(G) = max{w(G-S)-|S|-m(G-S): S \subset V(G), w(G-S)>1}

    where w(G-S) is the number of components of G-S and m(G-S) is the order of a largest component of G-S. In this paper, general results on the rupture degree of a graph are considered. Firstly, some bounds on the rupture degree are given. Further, rupture degree of a complete k-ary tree is calculated. Also several results are given about complete k-ary tree and graph operations. Finally, we give formulas for the rupture degree of the cartesian product of some special graphs.


  • [5] Mautner P., Mouček R. (Czech Republic): Processing and categorization of Czech written documents using neural networks, 53-66.

    Full text     DOI: 10.14311/NNW.2012.22.004

    The Kohonen Self-organizing Feature Map (SOM) has been developed for clustering input vectors and for projection of continuous high-dimensional signal to discrete low-dimensional space. The application area, where the map can be also used, is the processing of text documents. Within the project WEBSOM, some methods based on SOM have been developed. These methods are suitable either for text documents information retrieval or for organization of large document collections. All methods have been tested on collections of English and Finnish written documents. This article deals with the application of WEBSOM methods to Czech written documents collections. The basic principles of WEBSOM methods, transformation of text information into the real components feature vector and results of documents classification are described. The Carpenter-Grossberg ART-2 neural network, usually used for adaptive vector clustering, was also tested as a document categorization tool. The results achieved by using this network are also presented.


  • [6] Raja P., Pugazhenthi S. (India): On-line path planning for mobile robots in dynamic environments, 67-83.

    Full text     DOI: 10.14311/NNW.2012.22.005

    Motion planning of mobile robots is a complex problem. The complexity further increases when it comes to path planning in dynamic environments. This paper presents an algorithm for on-line path planning of mobile robots in unknown environments with moving obstacles. A mathematical model is established which considers all the current on-line information of robot as well as nearing obstacles. Particle Swarm Optimization technique is used to optimize the velocity parameters of the robot, to arrive at the shortest collision-free trajectory, satisfying dynamic constraints. Simulation results show that the proposed algorithm is computationally efficient and effective.